Synthesis of an intelligent adaptive algorithm for automated forecasting of daily electricity consumption

In recent years, the need for sufficiently accurate short-term forecasting of electricity consumption has sharply increased to enhance the efficiency of operational management and balancing of the power system in critical modes. Daily electricity consumption is a random numerical sequence containing statistical patterns suitable for prediction. This work synthesizes an intelligent adaptive linear algorithm for predicting daily electricity consumption in a district power grid (referred to as the forecaster). The intellectual capabilities of the proposed algorithm were demonstrated by its automatic detection of correlation patterns in the input stream of electricity consumption data, as well as cross-correlation dependencies between electricity consumption, air temperature, and the day of the week. The application of the developed forecaster also enabled the automatic utilization of identified data patterns to optimize the electricity consumption forecasting strategy, achieving the highest possible accuracy. As a result of synthesizing the intelligent adaptive algorithm, the forecasting errors became random, uncorrelated, and exhibited characteristics of a normal distribution. This indicates that all opportunities to improve forecasting accuracy had been exhausted (given the available data). During the analysis, significant non-stationarity in the forecasting process was identified. There are prolonged periods when the root mean square error of forecasting is very small, but there are also moments when forecasting accuracy deteriorates significantly due to unknown episodic disturbances. However, the forecaster corrects its gross random errors as quickly as possible. Over an entire year, the forecaster made a significant error on two consecutive days only once. In the self-learning process, the forecaster automatically optimized the numerical values of the 21 weighting factors. It is impossible to perform such work manually.